From 270f6d194c124294d7ca8256c6bf7b6010e8510c Mon Sep 17 00:00:00 2001
From: =?utf8?q?Christian=20K=C3=B6nig?= <deathsimple@vodafone.de>
Date: Tue, 14 Dec 2010 20:49:31 +0100
Subject: [PATCH] r600g: optimize temp register handling for LRP

---
 src/gallium/drivers/r600/r600_shader.c | 72 ++++++++++++++------------
 1 file changed, 38 insertions(+), 34 deletions(-)

diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 3deabbca3a9..e00c844a17b 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -864,19 +864,25 @@ static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx, struct r600_
 	return 0;
 }
 
-static int tgsi_op2_s(struct r600_shader_ctx *ctx, int swap)
+static int tgsi_last_instruction(unsigned writemask)
 {
-	struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
-	struct r600_bc_alu_src r600_src[3];
-	struct r600_bc_alu alu;
-	int i, j, r;
-	int lasti = 0;
+	int i, lasti = 0;
 
 	for (i = 0; i < 4; i++) {
-		if (inst->Dst[0].Register.WriteMask & (1 << i)) {
+		if (writemask & (1 << i)) {
 			lasti = i;
 		}
 	}
+	return lasti;
+}
+
+static int tgsi_op2_s(struct r600_shader_ctx *ctx, int swap)
+{
+	struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
+	struct r600_bc_alu_src r600_src[3];
+	struct r600_bc_alu alu;
+	int i, j, r;
+	int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
 
 	r = tgsi_split_constant(ctx, r600_src);
 	if (r)
@@ -1037,7 +1043,7 @@ static int tgsi_trig(struct r600_shader_ctx *ctx)
 	struct r600_bc_alu_src r600_src[3];
 	struct r600_bc_alu alu;
 	int i, r;
-	int lasti = 0;
+	int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
 
 	r = tgsi_setup_trig(ctx, r600_src);
 	if (r)
@@ -1057,10 +1063,6 @@ static int tgsi_trig(struct r600_shader_ctx *ctx)
 		return r;
 
 	/* replicate result */
-	for (i = 0; i < 4; i++) {
-		if (inst->Dst[0].Register.WriteMask & (1 << i))
-			lasti = i;
-	}
 	for (i = 0; i < lasti + 1; i++) {
 		if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
 			continue;
@@ -1587,13 +1589,7 @@ static int tgsi_op3(struct r600_shader_ctx *ctx)
 	struct r600_bc_alu_src r600_src[3];
 	struct r600_bc_alu alu;
 	int i, j, r;
-	int lasti = 0;
-
-	for (i = 0; i < 4; i++) {
-		if (inst->Dst[0].Register.WriteMask & (1 << i)) {
-			lasti = i;
-		}
-	}
+	int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
 
 	r = tgsi_split_constant(ctx, r600_src);
 	if (r)
@@ -1937,6 +1933,7 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
 	struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
 	struct r600_bc_alu_src r600_src[3];
 	struct r600_bc_alu alu;
+	int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
 	unsigned i;
 	int r;
 
@@ -1947,7 +1944,10 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
 	if (r)
 		return r;
 	/* 1 - src0 */
-	for (i = 0; i < 4; i++) {
+	for (i = 0; i < lasti + 1; i++) {
+		if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+			continue;
+
 		memset(&alu, 0, sizeof(struct r600_bc_alu));
 		alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD);
 		alu.src[0].sel = V_SQ_ALU_SRC_1;
@@ -1957,7 +1957,7 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
 		alu.src[1].neg = 1;
 		alu.dst.sel = ctx->temp_reg;
 		alu.dst.chan = i;
-		if (i == 3) {
+		if (i == lasti) {
 			alu.last = 1;
 		}
 		alu.dst.write = 1;
@@ -1970,7 +1970,10 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
 		return r;
 
 	/* (1 - src0) * src2 */
-	for (i = 0; i < 4; i++) {
+	for (i = 0; i < lasti + 1; i++) {
+		if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+			continue;
+
 		memset(&alu, 0, sizeof(struct r600_bc_alu));
 		alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL);
 		alu.src[0].sel = ctx->temp_reg;
@@ -1979,7 +1982,7 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
 		alu.src[1].chan = tgsi_chan(&inst->Src[2], i);
 		alu.dst.sel = ctx->temp_reg;
 		alu.dst.chan = i;
-		if (i == 3) {
+		if (i == lasti) {
 			alu.last = 1;
 		}
 		alu.dst.write = 1;
@@ -1992,7 +1995,10 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
 		return r;
 
 	/* src0 * src1 + (1 - src0) * src2 */
-	for (i = 0; i < 4; i++) {
+	for (i = 0; i < lasti + 1; i++) {
+		if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+			continue;
+
 		memset(&alu, 0, sizeof(struct r600_bc_alu));
 		alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD);
 		alu.is_op3 = 1;
@@ -2002,16 +2008,20 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
 		alu.src[1].chan = tgsi_chan(&inst->Src[1], i);
 		alu.src[2].sel = ctx->temp_reg;
 		alu.src[2].chan = i;
-		alu.dst.sel = ctx->temp_reg;
+
+		r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+		if (r)
+			return r;
+
 		alu.dst.chan = i;
-		if (i == 3) {
+		if (i == lasti) {
 			alu.last = 1;
 		}
 		r = r600_bc_add_alu(ctx->bc, &alu);
 		if (r)
 			return r;
 	}
-	return tgsi_helper_copy(ctx, inst);
+	return 0;
 }
 
 static int tgsi_cmp(struct r600_shader_ctx *ctx)
@@ -2020,13 +2030,7 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
 	struct r600_bc_alu_src r600_src[3];
 	struct r600_bc_alu alu;
 	int i, r;
-	int lasti = 0;
-
-	for (i = 0; i < 4; i++) {
-		if (inst->Dst[0].Register.WriteMask & (1 << i)) {
-			lasti = i;
-		}
-	}
+	int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
 
 	r = tgsi_split_constant(ctx, r600_src);
 	if (r)
-- 
2.30.2